Jev Vision

Python harness for GUI grounding: it divides a screenshot into grid tiles, has a vision model describe each tile neutrally, then asks typesafe-ai/jev to choose among the descriptions. Local confidence and margin gates decide whether to descend or refuse, and it emits only a point and bounding box.

Serves Shanghai AI Lab's Intern-Decision-0.8B vision decision model on Apple Silicon behind a /v1/systemone-shaped endpoint: screenshots plus typed questions in, calibrated Choice, Score and Noul probabilities out, for thresholding and escalation.

Multimodal finite-choice decision interface and MuJoCo robotic-arm workbench: dual-camera images and state go to a self-hosted Qwen3.5-9B with a corrective head, which returns one typed next-skill choice without generating text.

Jev-compatible decision models with vision input: LoRA fine-tunes of Qwen3.5 at 4B and 9B serving Choice, Score and Noul on the /v1/systemone wire format, with screenshots and photos placed beside text in the state. Runs on your own NVIDIA GPU; weights are non-commercial.

A multimodal decision model that scores text, image, and video candidates with a shared head on Qwen3.5, plus SFT and experimental RLCD training code.

Allan Riordan Boll reads one-token logprobs so a chat model returns a letter and a distribution, including on webcam frames from a vision model.